Interpreting Oaxaca Decomposition

[Boguslawa Stec]

Hello everyone,

I am using Oaxaca decomposition for my graduation project, where the subject is wage differences between gender, with some focus on difference between the public and private sectors. I first applied the decomposition method on wages in general to obtain the gender wage gap between genders. Later on, I have applied the threefold decomposition method for hourly wages in both public and private sectors to obtain a wage gap there. However, as can be observed in table below, in few of the years the "difference" and the "Interaction" has high p-values. What does this high value mean? Can I still make inferences about the gap? I can't find any sources that would discuss the p-values or their significance to the decomposition.

Public Sector
Variable	2002	2004	2006	2008	2010	2012	2014	2016
Men	2.12711*** (0.000)	2.127855*** (0.000)	2.28916*** (0.000)	2.326609*** (0.000)	2.372711*** (0.000)	2.389001*** (0.000)	2.432239*** (0.000)	2.41235*** (0.000)
Women	2.09901*** (0.000)	2.194716*** (0.000)	2.268984*** (0.000)	2.286633*** (0.000)	2.299852*** (0.000)	2.352422*** (0.000)	2.38888*** (0.000)	2.38355*** (0.000)
Difference	0.0281001 (0.284)	0.023138 (0.334)	0.020176 (0.491)	0.0399769* (0.099)	0.0728577*** (0.003)	0.0365788 (0.162)	0.0433593* (0.101)	0.0287993 (0.368)
endowments	-0.0390629** (0.047)	-0.0267182* (0.091)	-0.0492973** (0.018)	-0.0390105** (0.023)	-0.0403396** (0.019)	-0.0499241 *** (0.009)	-0.0683397*** (0.000)	-0.0665778*** (0.003)
Coefficients	0.0749937*** (0.003)	0.0551021** (0.029)	0.0830066*** (0.007)	0.06765505*** (0.006)	0.0802848*** (0.000)	0.0840823*** (0.000)	0.1038778*** (0.000)	0.0852044*** (0.003)
Interaction	-0.0078307 (0.670)	-0.0052451 (0.774)	-0.0135333 (0.554)	0.0113369 (0.527)	0.0329124** (0.021)	0.0024205 (0.888)	0.0078212 (0.646)	0.0101728 (0.597)
Explained	-0.0379897** (0.025)	-0.0302732* (0.054)	-0.0611286*** (0.000)	-0.0372746** (0.019)	-0.0188726 (0.241)	-0.0442742** (0.013)	-0.0646111*** (0.000)	-0.0712211*** (0.001)
Unexplained	0.0660897*** (0.003)	0.053412** (0.016)	0.0813046*** (0.003)	0.0772514*** (0.000)	0.0917303*** (0.000)	0.080853*** (0.000)	0.1079705*** (0.000)	0.1000204*** (0.000)
Number of obs	1565	1458	1316	1609	1559	1386	1027	706

All help will be greatly appreciated. Thank you.

[Marcos Almeida]

Maybe the difference itself is not significant for most years and the "issue" on significance is related to the (protective) effect for women with regard to the explained fraction

[Boguslawa Stec]

Thank you for your fast response Marcos. However, i am afraid i am not fully sure what you mean. Could you elaborate bit more?

[Boguslawa Stec]

Maybe the difference itself is not significant for most years and the "issue" on significance is related to the (protective) effect for women with regard to the explained fraction

Thank you for your fast response Marcos. However, i am afraid i am not fully sure what you mean. Could you elaborate bit more?

[Sven-Kristjan Bormann]

However, as can be observed in table below, in few of the years the "difference" and the "Interaction" has high p-values. What does this high value mean? Can I still make inferences about the gap?

Why do you think that you cannot make inferences about the gap when the p-values are high?
The high p-values just indicate that there might be no difference in the wages or that the endowment and coefficient effect move the difference to similar degree in opposite directions.
The high p-values for the interaction term indicates that differences in endowments and coefficients simultaneously between the two groups do not effect the difference in the outcome.
I admit, that for me the interaction term is the hardest to interpret. So, I usually treat it as some kind of "residual" term. The twofold decomposition is a bit easier to interpret in my opinion.

The p-values have the same meaning as in every other estimation or decomposition method. But maybe I misunderstand your problem with the p-values.

[Boguslawa Stec]

Why do you think that you cannot make inferences about the gap when the p-values are high?
The high p-values just indicate that there might be no difference in the wages or that the endowment and coefficient effect move the difference to similar degree in opposite directions.
The high p-values for the interaction term indicates that differences in endowments and coefficients simultaneously between the two groups do not effect the difference in the outcome.
I admit, that for me the interaction term is the hardest to interpret. So, I usually treat it as some kind of "residual" term. The twofold decomposition is a bit easier to interpret in my opinion.

The p-values have the same meaning as in every other estimation or decomposition method. But maybe I misunderstand your problem with the p-values.

Thank you for you answer Sven-Kristjan. I wasn't sure about the meaning behind the p-values, so it made me a little confused. Now thanks to your answer I understand what they mean and can now properly take on my analysis, I appreciate it.